Blaug, Capitulo 4

The falsificationists, a wholly twentieth-
century story
Ultraempiricism?
The year 1938 saw the publication of The Significance and Basic
Postulates of Economic Theory by Terence Hutchison, and with it the explicit
introduction of Popper's methodological criterion of falsifiability into eco-
nomic debates. That Hutchison should have recognized the significance of
Popper's demarcation criterion as early as 1938 is itself remarkable: Popper's
Logik der Forschung (1934) was then almost completely unknown, and even
so famous a popularization of the philosophical ideas of the Vienna circle as
Ayer's Language, Truth, and Logic (1936) completely missed the signifi-
cance of Popper's critique of the verifiability principle of meaning. To some
extent, even Hutchison failed to realize the novelty of Popper's thinking: al-
though he cited Popper frequently, he laid down the fundamental criterion
that economic propositions that aspire to the status of "science" must be
capable, at least conceivably, of being put to an interpersonal empirical test
without any acknowledgment to Popper (Hutchison, 1938, pp. 10, 19, 26-7,
48, 49, 126, 156).
22
Hutchison's principal target of attack was apriorism in
all its varieties, but in assailing the postulates of orthodox economics that
were said by Mises and Robbins to be intuitively obvious, he overstated his
case and so spoiled what might have been a decisive effort to reorient the
methodology of interwar economics.
At the center of Hutchison's argument is the notion that all economic prop-
22
When asked a few years later by Frank Knight to state his philosophical starting point,
it is significant that Hutchison (1941, p. 735) mentioned the British empiricists as well
as Mach, Schlick, and Carnap in Vienna without, however, referring to Popper. For
a later treatment of methodological issues in economics by a philosopher of the social
sciences that is simply Hutchison in different language, see Kaufmann (1944, chap.
16); he too makes no mention of Popper.
83
84 The falsificationists, a wholly twentieth-century story
ositions can be exhaustively classified into either tautological propositions or
empirical ones, the former being those that do not forbid any conceivable state
of the world and the latter those that do forbid at least some conceivable states
(1938, p. 13). Whatever we may think of such a dichotomous classification
of scientific propositions - some modern philosophers have questioned this
positivist dogma that all statements can be neatly divided into logically nec-
essary, "analytic" propositions and logically indeterminate, "synthetic" ones
(Nagel, 1961, p. 371) - the fact remains that Hutchison tended to characterize
most economic propositions as tautologies. In so doing, he blurred the vital
distinction in economics between assertions that are simply disguised defini-
tions and assertions that, while testable in principle, are so stated as to delib-
erately defy practical testing.
For example, metaphysical, "hard core" propositions in economics, such
as the belief that the price system invariably acts to harmonize the interests of
all economic agents or that all economic agents always act rationally in their
own interests, are indeed assertions about the real world but are nevertheless
irrefutable even in principle because they do not appear to forbid any events
from occurring. Similarly, Hutchison dismissed economic propositions with
unspecified ceteris paribus clauses as tautologies (1938, p. 42), whereas in
fact they are simply untestable empirical assertions about the real world. Con-
sider the two statements: the imposition of a tax on cigarettes will, ceteris
paribus, raise their price, and the imposition of a tax on cigarettes will, ceteris
paribus, lower their price; they cannot both be tautologies because they are
actually incompatible with each other. As they stand, both are "synthetic"
propositions about reality and yet neither is testable even in principle because
the cetera are not enumerated. Thus, if a statement is in principle falsifiable,
it does forbid some conceivable event or set of events. But the obverse does
not hold: a statement may forbid some conceivable set of events and yet be
irrefutable even in principle, as indeed are all tendency statements with un-
specified ceteris paribus clauses.
This criticism of Hutchison originates with Klappholz and Agassi (1967).
Instead of Hutchison's two-way classification into analytic-tautological prop-
ositions and synthetic-empirical ones, consigning most economic concepts to
the former category, Klappholz and Agassi propose a three-way classification
into (1) analytic-tautological propositions, (2) synthetic-empirical proposi-
tions that are nevertheless untestable even in principle, and (3) synthetic-
empirical propositions that are testable at least in principle, with the con-
sequence of reducing the number of economic concepts that fall into the
first category and increasing the number that fall into the second. Hutchison,
they contend, frequently criticizes economists for expressing tautologies
when in fact they are pronouncing untestable empirical assertions: "From
Ultraempiricism ? 85
his surveys of economic theory one gets the impression that most economic
theorists of his day uttered almost nothing but tautologies, although his
book appeared two years after Keynes's General Theory. Yet Keynes was
undoubtedly concerned with empirical issues" (Klappholz and Agassi, 1967,
p. 28).
23
Hutchison's principal methodological prescription is that scientific eco-
nomic inquiries should be confined to empirically testable statements. Unfor-
tunately, he is rather vague about the question of whether the requirement of
testability refers to the assumptions or to the predictions of economic theory.
On balance, he seems to emphasize the testing of postulates, or what we now
call assumptions, as suggested in the very title of his book, and this impres-
sion is strengthened by his response to Fritz Machlup's subsequent accusation
of ultraempiricism: Machlup (1978, pp. 143-4) cites Hutchison as a prime
example of an ultraempiricist, meaning one who would "insist on indepen-
dent verification of all assumptions by objective data obtained through sense
observation," thus proposing "a program that begins with facts rather than
assumptions." Hutchison (1956) denies the charge of ultraempiricism and has
no difficulty in showing that many of the statements in his book about the
importance of testability referred, not to the assumptions, but to "the finished
propositions" of economics.
Nevertheless, the burden of the book suggests otherwise and even the re-
joinder to Machlup, written almost twenty years after the book, contains hints
of Hutchison's long-standing conviction that empirical work in economics is
just as usefully applied to the assumptions as to the predictions of theory.
Thus, Machlup argues that the direct testing of such fundamental assumptions
as utility maximization on the part of households and profit maximization on
the part of business firms by, say, the interrogation of large numbers of con-
sumers and entrepreneurs is "gratuitous, if not misleading"; to this remark
Hutchison (1956, p. 48) replies: "it does not matter in principle whether the
specification of the conditions of a test of this fundamental assumption [of
rationality] is obtained 'directly' and 'independently', or by working back
'indirectly' from the specified tests of the conclusions to the assumptions from
which the conclusions are deduced." Actually, it matters a great deal and it
matters "in principle": it is precisely on this issue that Hutchison parts com-
pany with Machlup and, as we shall see, with Friedman's influential 1953
23
Hutchison was quite right in arguing that economists did (and do) protect substantive
empirical propositions by presenting them as if they were tautologies and definitions;
contrariwise, they sometimes manage miraculously to extract substantive economic
insights from what are in fact tautological identities (see Leontief, 1950; Klappholz
and Mishan, 1962; also Hutchison, 1960; Klappholz and Agassi, 1960; Hutchison,
1966; Latsis, 1972, pp. 239-41; Rosenberg, 1976, pp. 152-5).
''Essay on the Methodology of Positive Economics." Machlup is not far wrong
in labeling the Hutchison of 1956 and even more the Hutchison of 1938 as a
"reluctant ultra-empiricist" (Machlup, 1978, pp. 493-503).
Apriorism once again
If we are to do historical justice to Hutchison's book, however, we
need to remind ourselves once again of the strength of apriorism in the 1930s,
namely, the methodological view that economics is essentially a system of
pure deductions from a series of postulates derived from inner experience,
which are not themselves open to external verification. Thus, the publication
of Hutchison's book was greeted by a wild and confusing essay-length review
by Frank Knight, expressing profound irritation with what he took to be
Hutchison's "positivism," denying that truth in economics is anything like
truth in the natural sciences, affirming the Verstehen doctrine in economics,
24
and concluding: "It is not possible to 'verify' any proposition about 'econom-
ics' behaviour by any 'empirical' procedure, if the key words of this statement
are defined as they must be defined to be used with relevance and precision"
(Knight, 1956, p. 163; also pp. 164, 168). When Hutchison (1941) restated
his position, Knight came back with the categorical denial that propositions
about economic behavior can be empirically tested because economic behav-
ior is goal-directed and, therefore, depends for its meaning on our intuitive
knowledge of its purposive character:
My point was and is that the categorical contrast drawn by Mr. Hutchison, and so
many others [?], between propositions which can be tested and the ''value conceptions
of common sense", and the insistence that only propositions of the former character
are admissible in economic theory, is a false pretense and must simply be abandoned.
The testable facts are not really economics . . . This inability to test may or may not
be regarded as "too bad"; anyhow, it is the truth [Knight, 1941, p. 753; see also
Latsis, 1972, pp. 235-6].
It is curious that Knight, who in the early 1930s had become one of the
principal opponents of the Austrian theory of capital, should have continued
24
Similarly, Machlup (1978, pp. 152-3), in attacking Hutchison's ultraempiricism, de-
clares: "This, indeed, is the essential difference between the natural and the social
sciences: that in the latter the facts, the data of 'observation,' are themselves results
of interpretations of human actions by human actors. And this imposes on the social
sciences a requirement which does not exist in the natural sciences: that all types of
action that are used in the abstract models constructed for purposes of analysis be
'understandable' to most of us in the sense that we could conceive of sensible men
acting (sometimes at least) in the way postulated by the ideal type in question."
Operationalism 87
throughout his life to take his methodological views straight from Mises and
company (see Gonce, 1972; Hirsch and Hirsch, 1976, pp. 61- 5) .
It remains only to say that Hutchison in recent years has continued to insist
on the relevance of Popper's methodological prescriptions for economics, while
conceding that the advocacy of methodological monism is apt to be almost as
dangerous as the methodological dualism favored by advocates of Verstehen
doctrine.
Regarding the views expressed in that earlier essay [The Significance and Basic Pos-
tulates of Economic Theory}, I would still support for economics the criterion of test-
ability and falsifiability. However, though this earlier essay could be claimed to have
been, in many ways, a sceptical work by the standards of 1938, its optimistic "natu-
ralism" seems now indefensible: that is, its suggestions that the "social sciences"
could and would develop in the same manner as physics and the natural sciences . . .
It seems highly misleading to insist on certain general similarities between the natural
and social sciences (although such general similarities certainly exist), and to assert
that the differences are only ones "of degree," without making it clear how important
in practice these differences are [Hutchison, 1977, p. 151; see also pp. 57, 59-60;
and Hutchison, 1938, pp. vii-x].
25
Operationalism
In the same year that Ayer popularized logical positivism in Lan-
guage, Truth, and Logic, Percy Bridgman reaffirmed the methodology of
operationalism in The Nature of Physical Theory (1936). A year later, Paul
Samuelson began writing his Ph.D. thesis on the Foundations of Economic
Analysis, which carried the subtitle The Operational Significance of Economic
Theory. The thesis was finally published in 1948 and was immediately rec-
ognized as a major landmark in economic theory, not so much because of its
methodology, but because of its demonstration that the standard assumptions
of constrained maximization are not sufficient to derive most economic pre-
dictions: the method of comparative statics is empty unless a corresponding
dynamic system is specified and shown to be stable - this is the so-called
correspondence principle (Samuelson, 1948, pp. 262, 284).
One of the central purposes of his book, Samuelson declares, is to derive
4
'operationally meaningful theorems" in economics: "By a meaningful theo-
rem I mean simply a hypothesis about empirical data which could conceivably
be refuted if only under ideal conditions" (p. 4; also pp. 84, 91-2, 172, 220-
1, 257). Ironically enough, however, this is not operationalism as that term is
usually understood. The methodology of operationalism, as laid down by
Bridgman, is fundamentally concerned with the construction of certain cor-
25
For a review of Hutchison's entire career as a methodologist, see Coats (1983).
respondence rules that are supposed to connect the abstract concepts of a
scientific theory to the experimental operations of physical measurement. What
Samuelson's definition of operationally meaningful theorems amounts to,
however, is Popperian falsificationism expressed in the language of the Vi-
enna circle.
Samuelson goes on to draw a fundamental distinction in comparative static
reasoning between what has since come to be called the quantitative calculus
and the qualitative calculus. It is rarely possible in economics to specify the
magnitude of the change in the endogenous variables that results from a change
in one or more exogenous variables, but we must insist as a minimum require-
ment, Samuelson argues, that we can determine the algebraic sign of the
change: "The usefulness of our theory emerges from the fact that by our
analysis we are often able to determine the nature of the changes in our un-
known variables resulting from a designated change in one or more parame-
ters. In fact, our theory is meaningless in the operational sense unless it does
imply some restrictions upon observable quantities, by which it could con-
ceivably be refuted" (p. 7; also pp. 19, 21, 24ff, 257, 350-1). By applying
the criterion of the qualitative calculus to some of the main pillars of received
theory, Samuelson concludes that there is little empirical content in the mod-
ern theory of consumer behavior (pp. 90, 92, 97-8, 117, 172), and he is
equally skeptical about the principal tenets of the "new welfare economics"
which purport to make meaningful statements about welfare without resorting
to comparisons between individuals (pp. 244, 249).
The notion of an operationalist research program in economics has been
consistently ridiculed by Machlup. Reading Bridgman uncharitably (and
probably unfairly), Machlup interprets operationalism as ruling out all mental
constructs in theory formation, after which it is easy to show that this is
tantamount to eliminating all mathematical formulations of a theory. If, on
the other hand, we admit such mental operations as mathematical functions,
Machlup argues, the methodological force of operationalism is fatally com-
promised: theories made up only of operational concepts measurable in phys-
ical terms would amount to nothing more than lower-level generalizations
about empirical regularities (Machlup, 1978, chap. 6, especially pp. 179
83). This is so obvious that it would hardly be worth saying were it not for
the emotive impact of the adjective in the expression "operational theory,"
which, in Samuelson at any rate, is employed as a synonym for "empirical."
Machlup (1963, pp. 56-7) even goes so far as to deny that the concept of
equilibrium is deserving of the description "operational" - "Equilibrium as
a tool for theoretical analysis is not an operational concept; and attempts to
develop operational counterparts to the construct have not been successful"
- which seems to miss the significance of the qualitative calculus. The idea
of equilibrium is, surely, nothing more than the prediction that the real-world
Operationalism 89
observable counterparts of the endogenous variables of economic models will
remain constant so long as the real-world counterparts of the exogenous vari-
ables remain constant (Finger, 1971). In short, an operational theory is simply
a falsifiable one. Without mentioning the name of Samuelson, Machlup him-
self seems to imply as much when he says:
It is not easy to know what the economists who have used the phrase "operational
theory" really meant by this designation. They have not furnished any illustrations or
examples for their designation . . . What the economists could have meant when they
called for "operational theory" is . . . that a theory ought to have sufficient links with
the practical domain, with data of observation. Links are "sufficient" if they allow us
. . . to subject the theoretical system to occasional verification against empirical evi-
dence [1963, p. 66].
Precisely!
Donald Gordon (1955) makes a more promising effort to pin down the
meaning of operationalism in economics. He begins very much in the manner
of Bridgman by defining an operational proposition as one that states or im-
plies an operation that could in principle be performed, the results of which
would constitute a test of the proposition. But he allows the "operation" of
introspection in addition to the physical operations of recording, compiling,
and computing (1968, pp. 48-9) - just as Bridgman allowed pencil-and-
paper thought experiments - as a result of which his definition of operation-
alism is almost indistinguishable from Popper's definition of falsifiability. He
then applies the correspondence principle to reinterpret Samuelson's defini-
tion of operationally meaningful theorems: if a functional relationship among
observable variables is to have operational significance, the function must be
shown to be dynamically stable, that is, any departure from the equilibrium
solution of the endogenous variables motivates behavior that impels a return
to the original equilibrium solution; the test of the stability of a function is the
applicability of the qualitative calculus, implying in turn that the associated
ceteris paribus clause is subject to definite restrictions.
Thus, in the usual interpretation of ordinary demand curves, where we hold
constant the tastes of buyers as well as their incomes and the prices of closely
related commodities, the given incomes and prices are the cetera that restrict
the demand curve to certain empirically observable situations, whereas the
assumption of given tastes is an empirical hypothesis that the demand does
not shift, or shifts little, over the period of observation. It follows that, in
principle, there is no valid distinction between the quantitative and the qual-
itative calculus. If we can make qualitative predictions about the demand for
a product, it must be because its demand curve stays put over the period of
observation, in which case we may perhaps be able to predict its quantitative
90 Thefalsificationists, a wholly twentieth-century story
slope and elasticity. On the other hand, if we cannot make quantitative pre-
dictions about demand because the demand curve has shifted, we also cannot
derive qualitative predictions about the variations in demand. In practice,
however, the distinction between the quantitative calculus and qualitative cal-
culus is vital to the requirement of operational significance, or, as I would
prefer to say, to the requirement of falsifiability (Gordon, 1955, pp. 50-1).
The important principle that seems to be established by this argument is
that we can infer the existence of something like a well-defined, negatively
inclined demand function for butter (1) if we can correctly predict the alge-
braic sign of the change in the quantity of butter demanded consequent upon
a change in its price, and (2) if we can safely assume with the aid of the
correspondence principle that the butter market is dynamically stable. In the
Foundations, Samuelson frequently relies on casual empiricism to satisfy
condition (2), thus letting condition (1) do all the work in obtaining opera-
tionally meaningful theorems. To illustrate the point, consider the well-known
argument by which some teachers of first-year economics "prove" the prop-
osition that the marginal propensity to consume in a Keynesian macroeco-
nomic model must be less than unity: if it were equal to or greater than unity,
it follows by the definition of terms that the Keynesian multiplier would be
infinite, in which case the model would display the dynamic characteristic of
explosive instability; but the real world displays no such explosive instability;
therefore, the marginal propensity to consume must have a value of less than
unity. Q.E.D. Replying to Gordon in reference to all such arguments, Sam-
uelson (1966, pp. 1769-70) draws back from some of his earlier optimism in
the Foundations. The correspondence principle, he explains, is at best a heu-
ristic device and "In Foundations . . . , I stepped forward as a man of the
world and casual empiricist and stated my opinion that the hypothesis of dy-
namical stability was a 'realistic' one to make. I am no longer so sure of this
. . . your theoretical model or system will always be an idealized representa-
tion of the real world with many variables ignored; it may be precisely the
ignored variables that keep the real world stable."
The qualitative calculus and the correspondence principle have been further
developed and put to subsequent use in the testing of economic theories (e.g.,
Archibald, 1961; 1965; Lancaster, 1962, 1966a), but to say more of this now
is to run ahead of our story. At this point, we must turn to the centerpiece of
postwar economic methodology, the one essay on methodological questions
that virtually every modern economist has read at some stage in his or her
career: Milton Friedman's "Essay on the Methodology of Positive Econom-
ics" (1953). Its central thesis that economists should not bother to make their
assumptions "realistic" aroused a storm of controversy that took almost a
decade to die down,
26
and so subtle is Friedman's argument that even now it
26
So famous is Friedman's thesis that it has even become the subject of widely dissem-
The irrelevance-of-as sumptions thesis 91
is difficult to find two economists who will agree on precisely what it was that
Friedman said. In part, this is because the essay pursues two quite different
theses, which are presented as if one were the corollary of the other, though
they actually have very little to do with each other.
The irrelevance-of-assumptions thesis
Friedman opens his essay by setting out the old Senior-Cairnes-
Keynes distinction between normative and positive economics, after which
he asserts the essential methodological unity of all the physical and social
sciences, including economics in its positive part. There follows a statement
of the nature of that unified methodology (despite the Popperian ring of the
passage, it makes no explicit reference to Popper, or, for that matter, to any
other philosopher of science):
Viewed as a body of substantive hypotheses, theory is to be judged by its predictive
power for the class of phenomena which it is intended to "explain." Only factual
evidence can show whether it is "right" or "wrong" or, better, tentatively "ac-
cepted" as valid or "rejected." As I shall argue at greater length below, the only
relevant test of the validity of a hypothesis [notice that "only"] is comparison of its
predictions with experience. The hypothesis is rejected if its predictions are contra-
dicted ("frequently" or more often than predictions from an alternative hypothesis);
it is accepted if its predictions are not contradicted; great confidence is attached to it if
it has survived many opportunities for contradictions. Factual evidence can never
"prove" a hypothesis; it can only fail to disprove it, which is what we generally mean
when we say, somewhat inexactly, that the hypothesis has been "confirmed" by ex-
perience [Friedman, 1953, pp. 8-9].
From here, Friedman moves swiftly to his main target, namely, the notion
that conformity of the assumptions of a theory with reality provides a test of
validity different from, or additional to, the test of its predictions. This widely
held view, he writes, "i s fundamentally wrong and productive of much mis-
chi ef" (p. 14). Not only is it unnecessary for assumptions to be realistic, it is
a positive advantage if they are not: "t o be important . . . a hypothesis must
be descriptively false in its assumptions" (p. 14). This flamboyant exaggera-
tion is what Samuelson later christened ' ' the extreme version of the F-twist. ' '
It is far from clear, as many commentators have noted (Rotwein, 1959, pp.
inated jokes. O'Brien (1974, p. 3) says that students at Belfast University told him
the following story (I heard the same story told at a party of economists in Bangkok
four years earlier): "An economist, an engineer and a chemist were stranded together
on a desert island with a large tin of ham but no tin-opener. After various unsuccessful
exercises in applied science by the engineer and the chemist aimed at opening the tin,
they turned in irritation to the economist who all the while had been wearing a superior
smile. 'What would you do?' , they asked. 'Let us assume we have a tin-opener',
came the unruffled reply."
564-5; Melitz, 1965, pp. 40- 1; Nagel, 1961, pp. 42-4, 1968), what is meant
by ' 'realism" of assumptions. The assumptions of economic theory are some-
times said to be "unrealistic" in the sense of being abstract. As we have just
seen, this is certainly one of Friedman's meanings: "realistic" assumptions
are descriptively accurate in the sense that they take account of all the relevant
background variables and refuse to leave any of them out. Friedman has of
course no difficulty in showing that absolutely any theory that is not an exact
replica of reality idealizes the behavior of economic actors and oversimplifies
the assumed initial conditions and hence is descriptively inaccurate. Like-
wise, he also has no difficulty in showing that if simplicity is a desirable
criterion of good theory, all good theories idealize and oversimplify outra-
geously.
But there is another sense in which the assumptions of theories in a social
science like economics may be said to be "realistic," namely, whether they
ascribe motives to economic actors that we, fellow human beings, find com-
prehensible. Verstehen doctrine tells us that this is a desideratum of adequate
theorizing in the social sciences. Friedman in later portions of his essay leans
heavily on this interpretation of the phrase "realism of assumptions," and he
rejects it as categorically as he does the interpretation of descriptive accuracy:
whether businessmen testify that they strive to maximize returns, or even
whether they recognize the meaningfulness of the question being put to them,
is no test of the "realism" of what he calls "the maximization-of-returns
hypothesis" because a Darwinian process of competitive rivalry guarantees
that only those who actually maximize will survive. Under a wide range of
circumstances, he writes, "individuals behave as-if they were seeking ration-
ally to maximize their expected returns . . . and had full knowledge of the
data needed to succeed in this attempt" (p. 21). We may now rephrase Fried-
man to read: "to be important . . . a hypothesis must be descriptively false
in its assumptions" in the sense of imputing as-if motives to economic actors
that they could not possibly hold consciously (like assuming that billiard play-
ers calculate the angle and momentum of billiard balls every time they drive
a ball into a pocket); all that matters is whether the theory grounded on as-if
motives has predictive value. This is about as radical a rejection of Verstehen
doctrine as we could ask for and is tantamount to the methodology of instru-
mentalism: theories are only instruments for making predictions or, better
still, inference tickets that warrant the predictions that we make (Coddington
1972, pp. 12-13). Thus, the as-if formulation of economic hypotheses not
only refuses to offer any causal mechanism linking business behavior to the
maximization of returns; it positively rules out the possibility of such an ex-
planation.
But there is still a third sense in which the assumptions of theories may be
said to be "unrealistic," and it is perhaps this interpretation that most of
The irrelevance-of-assumptions thesis 93
Friedman's critics have had in mind. It is the case where the assumptions are
believed to be either false or highly improbable in the light of directly per-
ceived evidence about economic behavior (for example, when business firms
are observed to commit themselves to a fixed rule-of-thumb for pricing their
products irrespective of economic circumstances). However, while continu-
ing to deny the need to test assumptions directly, Friedman does allow for
"The Use of 'Assumptions' as an Indirect Test of a Theory," to cite the
heading of an important but frequently overlooked section of his essay (pp.
26-30). That is to say, the assumptions of one theory regarded as false on
grounds of casual empiricism may figure as the implications of a wider theory
whose consequences can be or have been tested, in which case these assump-
tions may be shown to be false in a particular domain but not in another.
This raises an important methodological point about the role of assumptions
in theorizing: it is, among other things, to specify the range of the intended
applications of a theory. As Friedman aptly observes: "the entirely valid use
of 'assumptions' in specifying the circumstances for which a theory holds is
frequently, and erroneously, interpreted to mean that the assumptions can be
used to determine the circumstances for which a theory holds" (p. 19). In
other words, we should not inspect the assumptions of the theory of perfect
competition to see whether it can be applied to the cigarette industry, because
if the theory is properly formulated, the circumstances under which it applies
are specified as among its essential components; we know before we begin
that the theory of perfect competition cannot apply to the highly concentrated
cigarette industry. Once we eliminate any reference to a theory's domain of
application, we render it untestable because every refutation can be countered
by the argument that it has been incorrectly applied. But, having introduced
this important methodological clarification, Friedman immediately spoils the
point by allowing the theory of perfect competition to apply to any firm what-
soever, depending on circumstances: "there is no inconsistency in regarding
the same firm as if it were a perfect competitor for one problem and a monop-
olist for another" (p. 36; also p. 42). In other words, he reverts once again to
an extreme instrumentalist interpretation of economic theories.
27
27
We must, surely, agree with Archibald (1963, pp. 69-70) when he conjectures, sup-
pose "we can successfully predict some of the behaviour of an economic unity from
theory A and some from theory B; where A is right B would be wrong, and vice
versa. One way of interpreting the situation is: 'different theories for different prob-
lems'. Another is: 'A and B are both refuted'. How do we now proceed? My view is
that the correct predictions of both A and B constitute part of our stock of useful
knowledge, available for what I call engineering purposes, but that both A and B are,
as scientific hypotheses, refuted. We might now endeavour to construct a more gen-
eral theory, incorporating A and B. Part of such a theory would be the specification
of the circumstances in which each subtheory would hold. Such a theory would be
susceptible of refutation, since the specification might be wrong. In the case of the
Having distinguished the three senses in which assumptions may be said to
be realistic or unrealistic, it must be added that Friedman considerably aggra-
vates the problem of gauging his meaning by writing throughout his essay
about "assumptions" in quotation marks without the slightest regard for the
different logical status of various kinds of assumptions. He does not even
explicitly distinguish between initial conditions, auxiliary hypotheses, and
boundary conditions. As Archibald (1959a, pp. 64-5) has pointed out, as-
sumptions in economics may refer to (1) statements of motivation such as
utility and profit maximization; (2) statements of overt behavior of economic
agents; (3) statements of the existence and stability of certain functional rela-
tionships; (4) restrictions on the range of variables to be taken into account;
and (5) boundary conditions under which the theory is held to apply. The
issue of realism of assumptions is clearly very different for each of these five
assumptions.
Likewise, Melitz (1965, p. 42) distinguishes between "auxiliary" assump-
tions, which are used in conjunction with a theoretical hypothesis in order to
deduce its logical consequences, and "generative" assumptions, which serve
to derive the hypothesis itself. Despite the fact that every assumption may
serve in either capacity, depending on the particular prediction in question,
some frequently employed assumptions in economics usually function in one
role rather than the other: ceteris paribus is typically an auxiliary assumption,
whereas profit maximization is typically a generative assumption. Although
the "realism" of both kinds of assumptions may be relevant, a discrepancy
between the auxiliary assumptions and reality is more serious for the test of a
theory than the lack of "realism" of the generative assumptions since the
latter are usually capable of a number of alternative interpretations. Suffice it
to say that the entire thesis of the irrelevance of assumptions has been bedev-
iled from the outset by the indiscriminate use of the term assumptions.
Machlup, coming to Friedman's rescue, distinguishes a whole class of as-
sumptions, postulates, or fundamental hypotheses: " 'heuristic principles'
(because they serve as useful guides in the analysis), 'basic postulates' (be-
cause they are not challenged for the time being), 'useful fictions' (because
they need not conform to the 'facts' but only be useful in 'as-if' reasoning),
'procedural rules' (because they are resolutions about the analytical proce-
dures to be followed), 'definitions assumptions' because they are treated like
purely analytical conventions)" (Machlup, 1978, p. 145; see also Musgrave,
1981). In any theory, these types of fundamental assumptions have to be
supplemented by what he calls "assumed conditions," that is, initial condi-
tions specified as to type of case, type of setting, and type of economy to
monopoly-competition mixture my complaint is precisely that it is an ad hoc mixture
and not a general theory which includes the appropriate specification, and it is there-
fore not susceptible of refutation."
The irrelevance-of-assumptions thesis 95
which the theory is to be applied and from which a deduced outcome is to be
inferred for purposes of testing (pp. 148-50). He agrees that to verify a theory
(he always speaks of verification rather than falsification), the "assumed con-
ditions" must correspond to observable situations, but he exempts all the
fundamental assumptions from such scrutiny. That consumers are able to rank
their preferences in a consistent order and that entrepreneurs prefer more profit
to less if each is equally risky are fundamental assumptions "which, though
empirically meaningful, require no independent empirical tests" (p. 147).
Not only are such direct, independent tests "gratuitous," Machlup adds, they
are even "misleading" because "the fundamental assumption [of maximiza-
tion] may be understood as an idealization with constructs so far removed
from operational concepts that contradiction by testimony is ruled out" (p.
147). That does not mean that it is inviolate, he grants, because it may be
rejected together with the theoretical system of which it is a part, if and when
a more satisfactory system is available.
In short, Machlup takes the view that a theory is never wholly discredited,
even in contexts where its fundamental assumptions are known to be false,
unless a better theory can be and perhaps has been offered. He concedes that
the assumption of consistent utility-maximizing and profit-maximizing behav-
ior are contrary to fact for some consumers and entrepreneurs (p. 498). The
problem as he sees it is that we cannot know how significant deviations from,
say, profit-maximizing conduct are except in the context of specific predic-
tions. Therefore, we ought "to accept maximizing conduct as a heuristic pos-
tulate and to bear in mind that the deduced consequences may sometimes be
considerably out of line with observed data. We can, to repeat, test empiri-
cally whether the outcome of people's actions is most of the time reasonably
close to what one would expect if people always acted as they are unrealisti-
cally assumed to act" (p. 498).
28
This divides the methodological arena, ac-
cording to Machlup, between extreme apriorists such as Mises, Knight, and
Robbins at one end and ultraempiricists such as Hutchison at the other end,
with the middle ground between these two extremes occupied by Zeuthen,
Samuelson, Lange, Friedman, and presumably himself: "none of them holds
that no conceivable kind of experience could ever cause him to give up his
28
Similarly, Bear and Orr (1967, p. 195), without endorsing the irrelevance-of-assump-
tions thesis, argue that assumptions are difficult to test in economics and hence it is
legitimate, as a second-best approach, to treat assumptions that do not flatly contradict
anything observable as correct and to proceed directly to a test of predictions. "It is
wrong categorically," they say, "to disregard predictions from a perfectly competi-
tive model on the ground that any of the four or five inadequately rationalized inter-
mediate textbook conditions of perfect competition do not hold. Such a rejection is
erroneous because of the difficulty in establishing how widely or how significantly the
actual situation varies from the perfect competition ideal, or, indeed of establishing
what an appropriate ideal of perfect competition may be. "
theory, and none of them wants his fundamental assumptions empirically tested
independently of the propositions with which they are combined when the
theory is applied" (p. 495).
The big bad wolf, therefore, is he who insists on the direct verification of
fundamental assumptions as the critical test of the validity of a theory in
advance of, or independently from, a test of its predictions. But was there
ever such a bad wolf? What the critics of Friedman have argued is (1) that
accurate predictions are not the only relevant test of the validity of a theory
and that if they were, it would be impossible to distinguish between genuine
and spurious correlations; (2) that direct evidence about assumptions is not
necessarily more difficult to obtain than data about market behavior used to
test predictions or, rather, that the results of examining assumptions are not
any more ambiguous than the results of testing predictions; (3) that the at-
tempt to test assumptions may yield important insights that help us to interpret
the results of predictive tests; and (4) that if predictive testing of theories with
patently counterfactual assumptions is indeed all that we can hope for, we
ought to demand that our theories be put to extremely severe tests.
29
To underline points (2) and (3), let us spend a moment on what is meant
by a "test" of assumptions. Now, it may be agreed that any attempt to inter-
rogate businessmen as to whether they seek to maximize profits, or to equate
marginal revenue and marginal costs, or to discount the returns from a capital
project at the cost of capital to the firm, is bound to produce ambiguous an-
swers whose interpretation will usually beg the very same question that is
being investigated. But other inquiries are possible: not "what are the objec-
tives of the firm?" but "what are the bits of information that are collected
before making strategic decisions?" or "how are such decisions actually made
and how are conflicts within the firm about strategic output and investment
decisions actually resolved?" The traditional theory of the firm treats the firm
as if it were a "black box" without explicating its internal decision-making
machinery. An inquiry that seeks to throw light on the nature of the "black
box" must, surely, illuminate the attempt to test the predictions of the black-
box theory of business behavior and, in any case, without such an inquiry the
predictions of the theory are almost as difficult to test as are the assumptions.
Amazingly enough, Friedman actually concedes this argument: asking
businessmen what they do and why they do it, he notes at one point in his
29
See Koopmans, 1957, p. 140; Archibald, 1959a, pp. 61-9; Rotwein, 1959, p. 556,
1973, pp. 373-4; Winter, 1962, p. 233; Cyert and Grunberg, 1963, pp. 302-8; Mel-
itz, 1965, p. 39; De Alessi, 1965; Klappholz and Agassi, 1967, pp. 29-33; Rivett,
1970, p. 137; McClelland, 1975, pp. 136-9; Coddington, 1976a; Rosenberg, 1976,
pp. 155-70; Naughton, 1978; in defense, see Machlup, 1978, p. 153n; Pope and
Pope, 1972 a, 1972b. For a survey of some of Friedman's critics, see Boland (1979),
which however gives the critics too little credit.
The F-twist 97
essay, is "almost entirely useless as a means of testing the validity of eco-
nomic hypotheses," although it may be useful in "suggesting leads to follow
in accounting for divergences between predicted and observed results"
(Friedman, 1953, p. 31n). So, apparently, the testing of motivational assump-
tions has some limited role to play in validating theories, point (1), and,
furthermore, it may prove productive in interpreting the results of predictive
tests, point (3), from which we may therefore infer point (2). Indeed, reread-
ing Friedman's essay we are struck by the fact that he is careful never to say
that the realism of assumptions is irrelevant without preceding it by the
adverb largely. In other words, he avoids the extreme versions of the
irrelevance-of-assumptions thesis, or what Samuelson has dubbed the F-twist.
The F-twist
The debate surrounding Friedman's essay was considerably confused
by Samuelson's attempt to reduce Friedman's argument to the "basic version
of the F-twist," in the course of which he dropped his earlier advocacy of
"operationalism" and instead opted for the methodology of "descriptivism,"
which left most of the combatants with the feeling that Friedman's method-
ology might be objectionable but Samuelson's new methodology was worse.
According to Samuelson, the F-twist comes in two versions: a basic ver-
sion, which asserts that the lack of realism of a theory's assumptions is rele-
vant to its validity, and an extreme version, which ascribes positive merit to
unrealistic assumptions on the grounds that a significant theory will always
account for complex reality by something simpler than itself. Ignoring the
extreme version, he concentrates his attack on the basic F-twist:
. . . is fundamentally wrong in thinking that unrealism in the sense of factual inaccu-
racy even to a tolerable degree of approximation is anything but a demerit for a theory
or hypothesis [1966, p. 1774].
. . . the doughnut of empirical correctness in a theory constitutes its worth, while
its hole of untruth constitutes its weakness. I regard it as a monstrous perversion of
science to claim that a theory is all the better for its shortcomings; and I notice that in
the luckier exact sciences, no one dreams of making such a claim [1972, p. 761].
But admitting that we ought to worry about factually inaccurate assumptions,
the real question is whether we should discard a theory merely because its
assumptions are known to be unrealistic. However, on that question Samuel-
son is silent. When we recall that even Friedman only asserted that unrealistic
assumptions are "largely" irrelevant for assessing the validity of a theory,
and adding the fact that many of the motivational assumptions of economic
theories involve directly unobservable variables, we are not actually any the
wiser as a result of Samuelson's vehement condemnation of the F-twist.
Samuelson goes so far as to supply a logical proof of the error of the
F-twist (1966, pp. 1775-6), but that proof presupposes a perfectly axioma-
tized ' 'Euclidean" theory whose wholly deductive structure ensures that as-
sumptions, theoretical propositions, and the consequences of those proposi-
tions all mutually imply one another. In fact, most economic theories are not
completely axiomatized and do not possess a simple logical structure, which
is precisely why there is some point in distinguishing the assumptions of the-
ories from their implications (see De Alessi, 1971, pp. 868-9; Machlup, 1978,
p. 481; Pope and Pope, 1972b, p. 236; Wong, 1973, p. 321). Moreover, even
a completely axiomatized theory that is in principle decomposable into its
assumptions cannot be empirically tested unless it is supplemented by initial
conditions and more or less "realistic" auxiliary assumptions that provide
measurable proxies for the analytical variables that appear in the theory. Thus,
the Samuelson proof of the fallacy of the F-twist seems to have reference only
to the formal role of theory as an analytical filing system for organizing our
ideas about reality and not to the substantive role of theory as an "explana-
tion" of reality (see McClelland, 1975, pp. 139-41; Rosenberg, 1976, pp.
170-2).
We would have thought that the weakest link in Friedman's argument is
his commitment to the methodology of instrumentalism. Once theories are
seen as nothing but instruments for generating predictions, the thesis of the
irrelevance of assumptions is irresistible. "The only relevant test of the valid-
ity of a hypothesis," Friedman tells us, "is comparison of its predictions with
experience." But such a comparison may show that a particular theory pre-
dicts extremely accurately although the theory as such provides no explana-
tion, in the sense of a casual mechanism, to account for the prediction. Sci-
ence, it might be argued, ought to do better than merely predict accurately.
But instead of questioning Friedman's implicit recourse to the symmetry the-
sis, Samuelson himself invokes the symmetry thesis by opting for the meth-
odology of descriptivism:
A Gallup poll count of the mail would seem to show there is a widespread will to
disbelieve in my rather hardboiled insistence upon "theory" as (strategically simpli-
fied) description of observable and refutable empirical regularities . . . a description
(equational or otherwise) that works to describe well a wide range of observable reality
is all the "explanation" we can ever get (or need desire) here on earth . . . An expla-
nation, as used legitimately in science, is a better kind of description and not some-
thing that ultimately goes beyond description [Samuelson, 1972, pp. 765-6; also 1966,
p. 1778].
Apart from the fact that the methodology of descriptivism is a little old-fashioned
(Nagel, 1961, pp. 118-29), we wonder what is the purpose of this strenuous
The Darwinian survival mechanism 99
insistence that the answer to the question "why?" is always an answer to the
question "how?" In the final analysis, Samuelson is almost as defensive about
economics as is Friedman.
Instrumentalism is an excessively modest methodology, and so is descrip-
tivism, which is simply a poor man's version of instrumentalism (Boland,
1970; Wong, 1973; Caldwell, 1980; but see Hammond, 1990). But apart from
excessive modesty, what is wrong with instrumentalism? Its weakness is that
of all black-box theorizing that makes predictions without being able to ex-
plain why the predictions work: the moment the predictions fail, the theory
has to be discarded in toto because it lacks an underlying structure of assump-
tions, an explanans that can be adjusted and improved to make better predic-
tions in the future. It is for this reason that scientists usually do worry when
the assumptions of their theories are blatantly unrealistic.
Both writers have been charged with saying the same thing in different
words. They have also been charged with failing to practice what they preach.
Machlup (1978, pp. 482-3) refers to Samuelson's international factor-price-
equalization theorem (see Chapter 11 below) to show that Samuelson is as
much an F-twister as Friedman, in the sense that he too infers apparently
significant real-world consequences from theoretical assumptions admitted to
be patently counterfactual. And Archibald (1961, 1963) has argued convinc-
ingly that Stigler and Friedman attack Chamberlin's theory of monopolistic
competition, not on the grounds of its poor predictive record, but on grounds
of consistency, simplicity, relevance, etcetera, that is, on the basis of the
theory's assumptions rather than its predictions. But waiving such debating
points, what is striking is that Friedman, Machlup, and Samuelson, each in
their own way, adopt what we have earlier called a defensive methodology
whose principal purpose seems to be to protect economics against the carping
criticism of unrealistic assumptions, on the one hand, and the strident demand
of severely tested predictions, on the other (Koopmans, 1957, pp. 141-2;
Latsis, 1976, p. 10; Diesing, 1985). We have dealt with the first half of this
defense, but we have as yet said nothing about the second half.
The Darwinian survival mechanism
Fritz Machlup, while urging the importance of empirical research in
economics, is nevertheless keen to underline the inconclusiveness of all tests
of economic hypotheses. We have already noted that he prefers the language
of verification to that of falsification, but he is perfectly aware of the Popper-
ian argument that verified theories are simply those that have so far resisted
falsification: "testing an empirical hypothesis results either in its disconfir-
mation or its non-disconfirmation, never in its definitive confirmation"
(Machlup, 1978, p. 140). With the aid of this terminological clarification, we
can now consider his skepticism about empirical testing in a field like eco-
nomics:
Where the economist's prediction is conditional, that is, based upon specified condi-
tions, but where it is not possible to check the fulfillment of all the conditions stipu-
lated, the underlying theory cannot be disconfirmed whatever the outcome observed.
Nor is it possible to disconfirm a theory where the prediction is made with a stated
probability value of less than 100 per cent; for if an event is predicted with, say, 70
per cent probability, any kind of outcome is consistent with the prediction. Only if the
same "case" were to occur hundreds of times could we verify the stated probability
by the frequency of "hi t s" and "misses." This does not mean complete frustration of
all attempts to verify our economic theory. But it does mean that the tests of most of
our theories will be more nearly of the character of illustrations than of verifications
of the kind possible in relation with repeatable controlled experiments or with recur-
ring fully-identified situations. And this implies that our tests cannot be convincing
enough to compel acceptance, even when a majority of reasonable men in the field
should be prepared to accept them as conclusive, and to approve the theory so tested
as "not disconfirmed," that is, as "O. K. " [p. 155].
This passage may be read as a perfectly valid criticism of "naive falsifica-
tionism," restating the Duhem-Quine thesis, but it may also be read as a plea
for still more "sophisticated falsificationism": it is precisely because tests of
economic theories are "more nearly of the character of illustrations than of
verifications" that we need as many illustrations as possible. But that implies
that economists should concentrate their intellectual resources on the task of
producing well-specified falsifiable predictions, that is, assigning less priority
to such standard criteria of appraisal as simplicity, elegance, and generality,
and more priority to such criteria as predictability and empirical fruitfulness.
It is fairly clear from the drift of Machlup's argument here and elsewhere,
however, that he would order his priorities precisely the other way round (see
Melitz, 1965, pp. 52-60; Rotwein, 1973, pp. 368-72). Throughout his long
career, in which he has returned repeatedly to the methodological problems
of economics, Machlup has been singularly ingenious in discounting all tests
of economic theories that critics have devised, but he has never stated what
evidence, if it materialized, he would be willing to regard as a refutation of,
say, the neoclassical theory of business behavior or the marginal productivity
theory of the demand for factors (e.g., Machlup, 1963, pp. 190, 207). There
is little point in commending empirical work, as he certainly does, if it never
really makes a difference to the beliefs one holds.
30
30
Machlup (1978, p. 46) has recently described himself as "a conventionalist - in the
sense of one who accepts as meaningful and useful basic propositions that make no
assertions but are conventions (resolutions, postulates) with regard to analytic proce-
dure."
Friedman's attitude to empirical testing is somewhat different from that of
Machlup: although he agrees that "There is never certainty in science, and
the evidence for or against a hypothesis can never be assessed completely
'objectively' " (Friedman, 1953, p. 30), he is convinced that the neoclassical
research program has been frequently tested and, moreover, that it has passed
most of these tests with flying colors. First of all, he argues, as we have seen,
that competition represents a Darwinian process that produces exactly the
same results that would ensue if all consumers maximized their utility and all
business firms maximized their profits, as a result of which the neoclassical
model predicts correctly even though its assumptions may be counterfactual.
(The classic statement of this argument is by Armen Alchian, and we will
therefore label it the Alchian thesis.) Furthermore,
An even more important body of evidence for the maximization-of-returns hypothesis
is experience from countless applications of the hypothesis to specific problems and
the repeated failure of its implications to be contradicted. This evidence is extremely
hard to document: it is scattered in numerous memorandums, articles and monographs
concerned primarily with specific concrete problems rather than with submitting the
hypothesis to test. Yet the continued use and acceptance of the hypothesis over a long
period, and the failure of any coherent, self-consistent alternative to be developed and
widely accepted, is strong indirect testimony to its worth [pp. 22-3].
This is without doubt the most frustrating passage in Friedman's entire
essay because it is unaccompanied by even a single instance of these "count-
less applications." No doubt, when the price of strawberries rises during a
dry summer, when an oil crisis is accompanied by a sharp rise in the price of
oil, when stock market prices tumble after the threat of a switch to a hard
money policy, we may take comfort in the fact that the implications of the
maximization-of-returns hypothesis have once again failed to be refuted.
However, given the multiplicity of hypotheses that could account for the same
phenomena, we can never be sure that the repeated failure to produce such
refutations is not a sign of the reluctance of economists to develop and test
unorthodox hypotheses. It would be far more convincing to be told what eco-
nomic events are excluded by the maximization-of-returns hypothesis, or bet-
ter still, what events, if they occurred, would impel us to abandon the hypoth-
esis. As Archibald (1959a, p. 62) has justly remarked, the true purport of the
passage about "countless applications" is "to encourage complacency and to
discourage that sceptical re-examination of the allegedly obvious that is the
prerequisite of progress." It suggests that Friedman, despite what he says
elsewhere, is not really interested in testing the maximization-of-returns hy-
pothesis and is instead seeking to confirm it. As we know, there is no hypoth-
esis so strange but that it is confirmed by evidence all around us. Besides, the
age of a maintained hypothesis and the absence of a widely accepted rival do
not provide "strong indirect testimony to its worth," to quote Friedman's
own words; every fallacious doctrine that was ever held has been defended on
such grounds.
There remains what I have labeled the Alchian thesis, that is, the notion
that all motivational assumptions in microeconomics may be construed as as-
if statements. This may be viewed as a knockdown version of the irrelevance-
of-assumptions thesis - it is pointless to debate the realism of as-if assump-
tions because such assumptions are by definition neither true nor false - or
else as a radical reinterpretation of the maximization-of-returns hypothesis
that in effect shifts the locus of rational action from the individual to the social
plane. By leaning heavily on the Alchian thesis, Friedman is in fact repudiat-
ing the methodological individualism that is commonly held to be embedded
in the neoclassical approach to economic questions: instead of deriving testa-
ble predictions in-the-large from the rational action of individual agents in-
the-small, the predictions of microeconomics are instead derived from a new
kind of causal mechanism, namely, a dynamic selection process that rewards
those businessmen who for whatever reasons act as if they were rational max-
imizers, while penalizing those who act in some other way by bankruptcy.
This is not a behaviorist reinterpretation of traditional theory but rather a new
theory. It is what I earlier referred to as Friedman's second methodological
thesis whose theoretical implications are so far-reaching that it is amazing
how widely it has been accepted and how little its special features have been
noticed (but see Koopmans, 1957, pp. 140-1; Archibald, 1959a, pp. 61-3;
Winter, 1962; Diesing, 1971, pp. 59-60, 299-303; Nelson and Winter, 1982,
pp. 139-44.)
31
The reference to a dynamic selection process shows immediately what is
wrong with the appeal to the Alchian thesis: traditional microeconomics is
largely, if not entirely, an analysis of timeless, comparative statics, and as
such it is strong on equilibrium outcomes but weak on the process whereby
equilibrium is attained. "Let the apparent immediate determinant of business
behavior be anything at all - habitual reaction, random choice, or whatnot,"
Friedman (1953, p. 22) tells us; "Whenever this determinant happens to lead
to behavior consistent with rational and informed maximization of returns,
31
Thus, Harry Johnson (1968, p. 5), endorsing the irrelevance-of-assumptions thesis,
states without qualification: "it has been shown . . . that whether firms consciously
seek to maximize profits and minimize costs or not, competition will eliminate the
inefficient firms; and that whether consumer behaviour is rational or purely random,
the demand curves for a product will tend to slope downwards as in the Marshallian
analysis. In consequence, it is possible for economists to treat the economy as an
interdependent system responding to change according to certain general principles of
a rational kind, with considerably more confidence than appeared justifiable thirty
years ago. " For other echoes of the Darwinian thesis, see Winter (1962, p. In).
the business will prosper and acquire resources with which to expand; and
whenever it does not, the business will tend to lose resources." But the pro-
cess whereby some firms prosper when their actual behavior approaches max-
imizing behavior takes time, and no reason has been given for believing that
those firms, having prospered in one period, will act consistently in the next
period; in other words, ''habitual reaction" may, but "random chance" cer-
tainly will not, result in any cumulative tendency for profitable firms to grow
relative to those that are not profitable. As Sidney Winter (1962, p. 240)
expresses it in his systematic critique of the Alchian thesis:
There is then a basic difficulty in the existing statements of the selection argument, a
difficulty which is rooted in the fact that the relative deviations from profit maximi-
zation of different firms may change through time. Since there has been no careful
treatment of the dynamic process by which some patterns of behavior are shown to be
viable and others nonviable, it has escaped notice that firms cannot in general be
unambiguously ranked in terms of their closeness to maximising behavior. Such a
ranking must, in general, presume a particular state of the environment, but the envi-
ronment is changed by the dynamic process itself.
To vindicate the Alchian thesis, we need to be able to predict behavior in
disequilibrium situations, that is, we need to supplement the standard theory
of the firm by a so far missing theory of entry and exit relating to the appear-
ance and disappearance of firms in the economic environment. Suppose there
are increasing returns to scale in production, or any other technologically
based cost advantages; if a nonmaximizing firm gains an initial advantage
over a maximizer, say, by entering the industry earlier in time, the scale
advantage may allow that nonmaximizer to grow faster than the maximizer
and to do so irreversibly; in consequence, the only firms that we observe are
firms that fail to maximize profits and that indeed carry "slack" (Winter,
1962, p. 243). Even the mere presence of differentiated products and associ-
ated advertising in an industry may produce a similar result. Now, of course,
we can define a set of assumptions - constant returns to scale, identical prod-
ucts, perfect capital markets, reinvestment of all profits, etcetera - that will
support the Alchian thesis, but that procedure will only bring us back full
circle to the question of the "realism" of assumptions (pp. 242-5). In a
nutshell, the problem with the Alchian thesis is the same as the problem of
reading progress into "the survival of the fittest" in Darwinian theory: to
survive, it is only necessary to be better adapted to the environment than one's
rivals, and we can no more establish from natural selection that surviving
species are perfect than we can establish from economic selection that surviv-
ing firms are profit maximizers. What is true of firms is true of techniques:
once a technique gets a head start, an entire industry may be locked into a
technique that is actually suboptimal. A beautiful example of this is the sur-
vival of the generally acknowledged nonoptimal typewriter keyboard with
which we are all familiar (David, 1985).
We sum up our long analysis of Friedman's essay by reiterating its three
central arguments, all of which combine to provide a sweeping warrant for
economists of all persuasions to build abstract models without excessive anx-
iety on the grounds of implausible assumptions: (1) assumptions are ' 'largely"
irrelevant to the validation of theories, which ought to be judged ''almost"
solely in terms of their instrumental value in generating accurate predictions;
(2) standard theory has an excellent predictive record as judged by "countless
applications . . . to specific problems"; and (3) the dynamics of competition
over time accounts for this splendid track record, whatever are the facts of
either overt behavior or the motivation for behavior on the part of individuals.
No wonder that Friedman's persuasively argued essay has been found ex-
tremely comforting to a whole generation of economists!
Looking back at the entire debate surrounding Friedman's essay, we cannot
help being struck by the lack of methodological sophistication that it dis-
played. The notion that theories can be neatly divided into their essential
components and that the empirical searchlight is to be directed solely at the
implications and never at any other parts of a theory can only be understood
as a reaction to a century of critical bombardment of orthodox theory, first by
the German historical school and subsequently by the American institution-
alists. The style of this criticism, which was invariably accompanied by the
crudest of objections to the assumptions of standard theory, paying absolutely
no attention to its predictive content, inevitably produced the reaction among
defenders of received doctrine that "assumptions are largely irrelevant." It is
as though generations of physicists had ridiculed Newton's theory of gravity
on the grounds that he committed himself to the patently unrealistic assump-
tion that the masses of moving bodies are concentrated at their center, which
might well have induced Newton to reply that predictions are everything and
assumptions nothing. Faced with the accusation that no theory with counter-
factual assumptions can be taken seriously, the thesis of the irrelevance of
assumptions is almost excusable.
Friedman has done major work in his long career in monetary economics,
macroeconomics, microeconomics, and welfare economics. In addition, he
has written as much popular economics as ten financial journalists added to-
gether. Much of that work does exemplify his methodological principles but
some of it contradicts it, say, by leaning heavily on plausibility of assump-
tions as a reason for believing certain economic theories. That may be because
while sounding superficially as if he were indebted to Popper, in fact he owes
more to John Dewey than to Karl Popper: he is a pragmatist rather than a
falsificationist. This point of view is argued persuasively in a fascinating re-
Naive versus sophisticated falsificationism 105
cent book by Hirsch and de Marchi (1990), a book which threatens to start
off a new round in the debate on "what Friedman really meant" in his essay
on the methodology of positive economics.
Naive versus sophisticated falsificationism
We are almost at the end of our story of explicit methodological
controversy in modern economics and the rest of it can be quickly told. The
late 1950s saw the publication of two books on economic methodology, both
of which denied that economics is a science. Sidney Schoeffler's study of The
Failures of Economics (1955) is reminiscent of a prewar book by Barbara
Wootton, Lament for Economics (1938), although it goes much further in
denying the scientific claims of economics. Schoeffler's central argument is
simplicity itself: the entire hypothetico-deductive tradition of economic theo-
rizing is a blind alley and economists must investigate the whole of the social
fabric, abandoning the pretension that there is such a thing as an autonomous
discipline of economics; scientific predictions are only possible when there
are universal laws unrestricted as to circumstances, and since the economic
system is always open to noneconomic forces and the play of chance, there
can be no economic laws and hence no economic predictions as such (Schoef-
fler, 1955, pp. 46, 162). All this is in fifty-four pages, after which the rest of
the book consists of a series of case studies of the failures of particular eco-
nomic models.
This wholly negative indictment is capped by a positive proposal for a new
kind of economics, which surprisingly enough turns out to be a general theory
of rational action based on inductive studies of decision making (pp. 189
221). There is little point in separating the sense from the nonsense in Schoef-
fler's argument (but see Klappholz and Agassi, 1967, pp. 35-8), because any
methodological prescription that amounts to wiping clean the entire slate of
received economics and to starting all over again from scratch may be dis-
missed out of hand as self-defeating: economists have always ignored and
will always continue to ignore the advice of those who claim that because one
cannot run, it is pointless to try to walk.
Andreas Papandreou's Economics as a Science (1958) employs a somewhat
different but equally extreme argument that turns on a distinction between
models and theories: for Papandreou, models, unlike theories, cannot be re-
futed because their relevant "social space" is not adequately characterized;
but even "basic theories" in economics have to be supplemented by auxiliary
assumptions, or "correspondence rules" that relate the theoretical variables
of the theory to the actual world to become the "augmented theories" that
are genuinely refutable. His indictment of current practice in economics is
simply that economists rarely formulate "augmented theories" and instead
are satisfied either with "models" or with "basic theories," which are vir-
tually irrefutable ex post explanatory schema (Papandreou, 1958, pp. 9-11,
136, 139, 144-5; also 1963).
In essence, Papandreou is making a case by generalizing the Duhem-Quine
thesis, which he somehow interprets as a peculiar difficulty of economic the-
ories (see pp. 134-5). Although he emphasizes the importance of "empirical
meaningfulness," he seems to confine "basic theories" to quantitative com-
parative statics and to deny that economics can boast at least some confirmed
qualitative predictions. But it is never easy to decide just what he does mean
because the entire argument is buried beneath formal mountains of a new set-
theoretical language for economics (see Klappholz and Agassi, 1967, pp. 33-
5; Rosenberg, 1976, pp. 172-7). Papandreou's strident positivism appears to
have spawned a disciple who applied the essentials of the argument to the
theory of consumer behavior (Clarkson, 1963), but of that more anon (see
Chapter 6 below).
The next item in our chronology is Joan Robinson's Economic Philosophy
(1962), a puzzling little book that depicts economics as partly a scientific
study of society and partly a vehicle for propagating ideology, that is, special
pleading of a politically apologetic kind, but whose cumulative impact is to
suggest that received economics is much more the latter than the former.
Popper is mentioned as demarcating a metaphysical proposition from a sci-
entific one, and the inherent difficulties in social science of producing clinch-
ing evidence for theories are given as the reason that ideology so frequently
creeps into the argument: "economics limps along with one foot in untested
hypotheses and the other in untestable slogans" (Robinson, 1962, p. 25; also
pp. 3, 22-3). The book ends with a plea for not abandoning "the hope that
economics can make an advance towards science" (p. 146), but no guidance
is offered as to how this is to be achieved.
This brings us to the first edition of Richard Lipsey's popular textbook, An
Introduction to Positive Economics (1963), whose opening chapter on scien-
tific method amounted to a frank espousal of Popperian falsificationism in its
"naive" version, namely, the belief that scientific theories can be faulted by
a single decisive test. "Naive falsificationism" in the first edition gave way
to "sophisticated falsificationism" in the second edition: "I have abandoned
the Popperian notion of refutation and have gone over to a statistical view of
testing that accepts that neither refutation nor confirmation can ever be final,
and that all we can hope to do is to discover on the basis of finite amounts of
imperfect knowledge what is the balance of probabilities between competing
hypotheses" (Lipsey, 1966, p. xx; see also p. 52n).
32
The viewpoint that this
32
The source of this volte-face was events at the London School of Economics, where
Lipsey taught at the time, in the years from 1957 to 1963, a story well told by de
Marchi (1988).
Back to essentialism 107
passage exemplifies appears in all the subsequent editions of the book, and to
this day Lipsey's textbook remains the outstanding Popper-inspired introduc-
tion to elementary economics, which continually emphasizes throughout all
its pages the need to assess empirical evidence in favor of a particular theory
relative to the evidence supporting rival theories.
Back to essentialism
At this point, we may be tempted to echo Hutchison's recently ex-
pressed opinion that by now "Perhaps a majority of economists - but not all
- would agree that improved predictions of economic behaviour or events is
the main or primary task of the economist" (Hutchison, 1977, p. 8). It is
never easy to assess the balance of opinion on a matter such as this but suffice
it to say that there are plenty of indications that the majority, if it is a majority,
represents no more than 51 percent of modern economists. Radical econo-
mists, Marxists and neo-Marxists, post-Keynesian and neo-Keynesians, insti-
tutionalists, and heterodox economists of all kinds, who together constitute a
sizeable portion of the younger generation, would certainly not agree that
economic theories must ultimately stand or fall on the basis of their predic-
tions, or that empirical testing of hypotheses constitutes, as it were, the Mecca
of the modern economist (see Blaug, 1990, pp. 60-2). Even Benjamin Ward's
aggressive catalog of Whats Wrong With Economics? (1972), one of the best
efforts to date to reassess economics through Kuhnian spectacles, denies that
the failure to emphasize the empirically falsifiable consequences of theories
is one of the major flaws of modern economics (Ward, 1972, p. 173).
To show how far an anti-Popperian methodology actually prevails in some
quarters of the profession, we need only examine a radical methodological
contribution by Martin Hollis and Edward Nell, Rational Economic Man,
subtitled A Philosophical Critique ofNeo-Classical Economics (1975).
This book examines the unholy alliance between neoclassical economics
and logical positivism, without however mentioning either Popper, Lakatos,
or any other positivist later than the young Ayer (some of Popper's works are
cited in the bibliography but no explicit or implicit reference to his ideas
appears in the text). Positivism, they argue, is a false philosophy and neo-
classical economics must fall with it: the positivist thesis of the separability
of facts and values, on the one hand, and facts and theories, on the other, is
untenable because all facts are theory-laden and all theories are value-laden.
A more satisfactory epistemology can be built on rationalism by which they
mean the demonstration that there are Kantian "synthetic" a priori truths:
"Our strategy depends on being able to pick out what is essential and then to
insist that what is essential is therefore to be found in practice" (Hollis and
Nell, 1975, p. 254; also p. 178). Economic systems must reproduce them-
selves and this fact of reproduction is therefore the "essence" of economic
systems that alone can furnish a sound basis for economic theory. The trouble
with neoclassical economics, they say, is that there is nothing in the frame-
work that guarantees that firms and households will reconstitute themselves
on a period-to-period basis.
After this, we might have expected to learn that "sound" economic theory
is modern growth theory, which is of course fundamentally concerned with
the infinitely reproducible, steady-state properties of economic growth paths.
But no, the only alternative to neoclassical economics that incorporates the
essential aspect of "reproduction" is classical-Marxian economics, meaning
actually neo-Ricardian economics leaning heavily on the work of Sraffa rather
than Marx (Hollis and Nell, 1975, pp. 188, 195). The concluding chapter of
the book, giving a brief sketch of "Classical-Marxian Economics on Ration-
alist Foundations," seems to retract much of what has gone before: suddenly
recalling that capitalism is subject to periodic business cycles and perhaps
even to final collapse, the authors concede that "systems in fact often fail to
reproduce themselves," in which case it is difficult to see why so much was
made of "reproduction" as the essence of economic problems.
Hollis and Nell try to saddle "positive economists" with the problem of
induction; by demolishing induction they believe that they have scotched any
notion of a fruitful neoclassical research program. They inveigh against the
typical assumptions of neoclassical economics, particularly the perfect infor-
mation assumptions, in apparent ignorance of Hutchison who, as early as
1938, made many of their points, and they stress various genuine difficulties
in the attempt to test economic theories as if no one before them had ever
suspected such problems. In some mysterious sense, classical-Marxian eco-
nomics is supposed to escape all these difficulties, but of course it does so
only by evading the empirical yardstick for validating theories. Indeed, it is
clear that their rationalistic, essentialist approach to economic knowledge leaves
no role whatever for quantitative-empirical research. Their book simply wipes
out all the advances in methodological thinking in postwar economics that
Popperianism ushered in. We might almost say that if they had read Popper's
many devastating comments on the philosophy of essentialism (Popper, 1957,
pp. 26-34; also Popper, 1976, pp. 18-21, 197-8; Popper and Eccles, 1977,
pp. 172-94), their book would have been deprived of its raison d'etre.
This is perhaps as good a place as any to say a few more words about the
philosophy of essentialism, which will raise its ugly head once or twice more
in the course of our discussion. Essentialism goes back to Plato and Aristotle
for whom knowledge or "science" begins with observations of individual
events and proceeds by simple inductive enumeration until grasping by intui-
tion that which is universal in the events - their "essence" - which is then
enshrined in a definition of the phenomenon in question. The doctrine that it
is the aim of science to discover the true nature or essence of things and to
Institutionalism and pattern modeling 109
describe them by means of definitions had an enormous influence on Western
thought right up to the nineteenth century. Popper contrasts this brand of
methodological essentialism with the methodological nominalism that came
into scientific debates with Newton, according to which the aim of science is
to describe how things behave in various circumstances with the aid of uni-
versal laws, and not to determine what they really are.
Popper has long argued that essentialism has damaging effects on social
theories because it encourages an antiempirical tendency to solve problems
by the use of definitions. Hollis and Nell never in fact tell us how to go about
selecting the "essence" of economic systems; they imply that it amounts to
abstracting "correctly" but they provide no criterion for assessing "correct"
abstraction other than crude realism.
33
Adherents of essentialism are inclined
to settle substantive questions by reaching for a dictionary of their own mak-
ing, and Hollis and Nell exemplify this tendency to perfection: reproduction
is the "essence" of economic systems because we tell you so!
Institutionalism and pattern modeling
Have I covered the entire menu of possible economic methodolo-
gies? Some would say not. They discern in the writings of the American
institutionalists a mode of explanation that is neither apriorism, convention-
alism, operationalism, instrumentalism, descriptivism, nor falsificationism: it
is what has been called pattern modeling because it seeks to explain events or
actions by identifying their place in a pattern of relationships that is said to
characterize the economic system as a whole (Wilber and Harrison, 1978).
Pattern modelers, we are told, reject all forms of "atomism" and refuse to
abstract from any part of the whole system; their working hypotheses are
relatively concrete and close to the system being described, and if they gen-
eralize at all, they do so by developing typologies; their explanations empha-
size "understanding" rather than "predictions," and they view an explana-
tion as contributing to understanding if new data fall into place according to
the stated patterns.
I have no doubt that this is a more or less accurate description of the meth-
ods of some institutionalists such as Thorstein Veblen, Clarence Ayers, and
perhaps Gunnar Myrdal. But it is difficult to find anything like pattern mod-
eling in the writings of John R. Commons, Wesley Clair Mitchell, and John
Kenneth Galbraith, whom some would regard as leading institutionalists. It is
clear that all of these writers are united in some respects: none of them will
have any truck with concepts of equilibrium, rational behavior, instantaneous
33
Thus, Nell (1972a, p. 94) writes elsewhere, "we must examine the definitions and
assumptions of our models for their realism, and for the extent to which they incor-
porate the essentials. If they are realistic, the working of the model should mirror that
of the economic system in a relatively simple and abstract form."
adjustments, and perfect knowledge, and they all favor the idea of group
behavior under the influence of custom and habit, preferring to view the eco-
nomic system more as a biological organism than as a machine. But that is a
far cry from saying that they share a common methodology, that is, a common
method of validating their explanations (see Blaug, 1978, pp. 710-13, 726-
7). There may be such a thing as a school of institutionalism, but it clearly
has no unique methodology denied to orthodox economists.
A much better description of the working methodology of institutionalists
is what Ward (1972, chap. 12) labels storytelling, which he argues also de-
scribes much orthodox economics, particularly of the applied kind. Storytell-
ing makes use of the method of what historians call colligation, the binding
together of facts, low-level generalizations, high-level theories, and value
judgments in a coherent narrative, held together by a glue of an implicit set
of beliefs and attitudes that the author shares with his readers. In able hands,
it can be extremely persuasive, and yet it is never easy to explain afterwards
why it has persuaded.
How does one validate a particular piece of storytelling? One asks, of course,
if the facts are correctly stated; if other facts are omitted; if the lower-level
generalizations are subject to counterexamples; and if we can find competing
stories that will fit the facts. In short, we go through a process that is identical
to the one that we regularly employ to validate the hypothetico-deductive
explanations of orthodox economics. However, because storytelling lacks rigor,
lacks a definite logical structure, it is all too easy to verify and virtually im-
possible to falsify. It is or can be persuasive precisely because it never runs
the risk of being wrong.
Perhaps economic problems are so intractable that storytelling is the best
that we can do. But if such is the case, it is odd that we should actually
recommend the safe methodology of storytelling and deplore the risky meth-
odology of falsificationism. Surely, the more falsificationism, the better?
The current mainstream
Nothing like an overwhelming consensus has emerged from our sur-
vey of postwar economic methodology. But despite some blurring around the
edges, it is possible to discern something like a mainstream view. Despite the
embarrassment of the F-twist, Friedman and Machlup do seem to have per-
suaded most of their colleagues that direct verification of the postulates or
assumptions of economic theory is both unnecessary and misleading; eco-
nomic theories should be judged in the final analysis by their implications for
the phenomena that they are designed to explain. At the same time, econom-
ics is held to be only a "box of tools," and empirical testing can show, not
so much whether particular models are true or false, but whether or not they
are applicable in a given situation. The prevailing methodological mood is not
The current mainstream 111
only highly protective of received economic theory, it is also ultrapermissive
within the limits of the "rules of the game": almost any model will do pro-
vided it is rigorously formulated, elegantly constructed, and promising of
potential relevance of real-world situations. Some famous economists, like
the late John Hicks for example, even manage at one and the same time to
pooh-pooh empirical testing and to emphasize the policy implications of eco-
nomic theories, a patently schizophrenic position (see Blaug, 1990, chap. 5).
Modern economists frequently preach falsificationism, as we have seen, but
they rarely practice it: their working philosophy of science is aptly described
as "innocuous falsificationism."
34
To substantiate this charge, we will examine the empirical status of a se-
lected sample of ruling economic theories. Before we do so, however, we
need to digress to consider the troublesome question of welfare economics.
One of the features that distinguishes economics from physics, chemistry, and
biology is that propositions in economics frequently serve at one and the same
time as explanations of behavior and as stipulated norms for behavior. There
is little or nothing in the modern philosophy of science that helps us to judge
theories that deduce the nature of a social optimum from certain fundamental
value judgments. Is this perhaps why so many modern economists fail to take
falsificationism seriously?
34
I owe this happy phrase to Coddington (1975, p. 542).

Blaug, Capitulo 4

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Blaug, Capitulo 4

Uploaded by

Copyright:

Available Formats

The falsificationists, a wholly twentieth-

You might also like